Author: Bruna Wundervald License: MIT

hebart: A package for fitting of Hierachical Embedded Bayesian Additive Regression Trees, which is an extension of BART to grouped data.

hebart description goes here

Installation

You can currently install the latest version of hebart from the github repository with:

# install.packages("devtools")
devtools::install_github("brunaw/hebartBase")

This package wil soon be on CRAN;

install.packages("hebart")

Functionalities

All of the functions and documentation can be found with:

library(hebart)
packageVersion("hebart")
ls("package:hebart")
help(package = "hebart")

Examples

# Package loading  ----------------------------------
library(magrittr)
library(ggplot2)
library(tidymodels)
library(hebartBase)

# Dataset split  ------------------------------------
set.seed(2022)
df_real     <- lme4::sleepstudy %>% set_names(c('y', 'X1', 'group'))
df_real$y   <- c(scale(df_real$y))
data_split  <- initial_split(df_real)
train       <- training(data_split)
test        <- testing(data_split)

# Running the model ----------------------------------
hb_model <- hebart(formula = y ~ X1,
                   data = train,
                   group_variable = "group", 
                   num_trees = num_trees,
                   priors = list(
                     alpha = 0.95, # Prior control list
                     beta = 2,
                     nu = 2,
                     lambda = 0.1,
                     tau_mu = 16 * num_trees,
                     shape_sigma_phi = 0.5,
                     scale_sigma_phi = 1,
                     sample_sigma_phi = TRUE
                   ), 
                   inits = list(tau = 1,
                                sigma_phi = 1),
                   MCMC = list(iter = 500, 
                               burn = 250, 
                               thin = 1,
                               sigma_phi_sd = 0.5)
)

hb_model

Hebart result
-----------------------------------
Formula:
 y ~ X1 

Number of trees:         10 
Number of covariates:    1 
Training error (RMSE):    0.3851619 
R Squared:                0.842164 


# Making predictions ----------------------------------
pred_test <- predict_hebart(newX = test, new_groups = test$group,
                            hebart_posterior  = hb_model, type = "mean")

# Predictions in the test set --------------------------
data.frame(
  y = test$y, pred = pred_test, 
  group = test$group 
)  %>% 
  ggplot(aes(x = y, y = pred)) +
  geom_point(aes(colour = factor(group)), size = 2) +
  labs(x = "True y", y = "Predictions") +
  theme_light(16) +
  scale_y_continuous(breaks = scales::pretty_breaks(n = 10)) +
  scale_x_continuous(breaks = scales::pretty_breaks(n = 10)) +
  guides(colour = "none")
 

# Diagnostics --------------------------
diagnostics(hb_model)
#----------------------------------------------------

To be implemented

• The official pkgdown for hebart
• Variable importances

Citation

To cite this package in publications, please use:

Bruna Wundervald (2022). hebartBase: Hierachical Embedded Bayesian Additive Regression Trees. R package version 0.1.0.
https://CRAN.R-project.org/package=hebartBase 

A BibTeX entry for LaTeX users is

@Manual{,
  title = {hebartBase: Hierachical Embedded Bayesian Additive Regression Trees},
  author = {Bruna Wundervald},
  year = {2022},
  note = {R package version 0.1.0},
  url = {https://CRAN.R-project.org/package=hebartBase},
}

This citation format can be obtained at any moment in R with:

citation('hebart')

Contributing

Contributions to this project are always highly incentivized. To do so, please be aware that git is our main tool for version control. The minimal steps for a contribution are:

1. Fork this repository into your GitHub account and clone it the way you prefer.
2. Do the changes, making sure everything is well documented, examples are provided and checking if the package still correctly builds.
3. Push your changes to git and create a new pull request in GitHub, explaining why and what are the changes made.
4. Done! Wait for review & acceptance of the pull request :)

To contributors who are new to writing R packages, we recommend the 'R Packages' book, by Hadley Wickham. To those who are new to git/GitHub, we recommend this tutorial. Many contributing resources to open source projects can be found at this repository.